NVIDIA, TSMC hold pricing power

Semiconductors are turning into a bottleneck business again. Not the old kind, where everyone gets crushed in a downturn, but the opposite — a few companies control the scarce pieces everyone suddenly needs, and they can keep charging more. That is the real story behind the latest numbers from TSMC and SK Hynix. In April, both companies showed that AI demand is still strong enough to preserve pricing power even after two years of nonstop buildout. The bigger point is macro — if the most important inputs for AI systems keep getting pricier, that cost does not stay inside the chip sector. It spreads. ### Which companies actually have the leverage? Nvidia has the system-level leverage. TSMC has the manufacturing leverage. SK Hynix, Micron, and Samsung have the memory leverage. Those are the choke points in AI hardware right now — accelerators, advanced foundry capacity, advanced packaging, and high-bandwidth memory. There are substitutes at the edges, but not many where performance really matters. ### What changed in the latest results? TSMC reported first-quarter net income of NT$572.48 billion, up 58% from a year earlier, and guided to more than 30% revenue growth for 2026. Management also pushed capital spending toward the upper end of a range that goes as high as $56 billion — basically saying demand is strong enough to justify another huge expansion cycle. ### Why is TSMC so central? Because the constraint is no longer just leading-edge wafers. It is also packaging — especially CoWoS, the advanced packaging used to stitch together AI processors and memory. When that step is scarce, TSMC is not just a manufacturer. It is the tollbooth. Even aggressive capacity expansion tells you the same thing: customers still need more than the system can comfortably supply. (cnbc.com) ### What is happening in memory? This is where the pricing story gets loud. SK Hynix posted record first-quarter revenue and profit, with operating margin hitting 72%, and said customers are prioritizing procurement over price. That is about as clear a pricing-power signal as you get. TrendForce then put numbers on the squeeze — conventional DRAM contract prices are projected to rise 58% to 63% quarter over quarter in Q2 2026, while NAND prices are seen rising 70% to 75%. (bloomberg.com) ### Why would ordinary hardware buyers care? Because capacity is being pulled toward the most profitable AI products first. Suppliers are reallocating output toward HBM, server DRAM, and enterprise storage. That leaves less room for PCs, phones, graphics memory, and lower-margin components. So even buyers that are nowhere near a giant AI cluster can still feel the squeeze through higher server, laptop, or device input costs. (cnbc.com) ### Does this really connect to inflation? Not in a clean one-to-one way — but yes, that is the concern. Investors have already been flagging AI buildout as a source of higher chip and power costs, which could keep broader inflation stickier than markets expect. Cleveland Fed data updated May 12 shows 10-year expected inflation at 2.48%, not a panic number, but still a reminder that inflation expectations have not vanished. (trendforce.com) ### So what is the real takeaway? The AI boom is no longer just a demand story. It is a supplier-power story. Nvidia, TSMC, and the top memory makers sit at the narrowest points in the chain, and narrow points set prices. Until capacity catches up — and that still looks a ways off — AI keeps looking less like a deflationary tech wave and more like an expensive industrial buildout with real pricing spillovers. (srnnews.com)